Analyzing dataset features:

A. FlightPhase, Altitude - Features¶

Visualize the relationship between Altitude and FlightPhase

Countplot

Observations:¶

• Over 10000 incidents happened during Approach, which is when the aircraft aligns with the runway and prepares to land.
• Above 4000 incidents took place at each of the following phases: Climb, Landing Roll, and Take-off run.
• Although not many, incidents happened during the Parking and Taxi phases.

Plot the Heatmap¶

Visualize count in Flight Phase and Altitude levels (both are linked)¶

Show how Damage is distributed across Flight Phase and Altitude levels¶

Pie chart displaying the proportion of incidents at each Flight Phase¶

Observations:¶

1. The data shows that most of the incidents take place during the Take-off run phase, compared to the remaining phases. Although most of these incidents caused no damage to the aircraft.
2. Almost equal number of incidents take place during Climb, Taxi, and Approach.
3. During the Descent phase, fewer incidents occurred, maybe because the airplane didn't find many birds at that altitude. But, during this phase, more incidents resulted in damage.
4. The number of incidents in each phase corresponds to the altitude at which birds are most likely to be present.

Correlation¶

Data shows that most incidents took place during Approach, although the vast majoprity did not caused damages.

Trend¶

The trend suggests that more incidents occurred during the Approach phase, but these incidents were generally less severe and caused no damage. On the other hand, fewer incidents took place during the Descent phase, but there was a higher likelihood of these incidents resulting in damage.

Note: This could be due to the nature of the incidents or the fact that there are fewer obstacles like birds at lower altitudes during descent.

B. Pilots Warned - Feature¶

Explore the counts of PilotWarned and Damage

Observations:¶

1. The charts above confirm the expected trend: incidents where pilots were warned caused less damage compared to those where pilots were not warned.
2. However, the instances in which pilots were warned account for only 15% more of the total incidents.

Correlation¶

For incidents that did not cause damage, the number of incidents when the pilot was warned was significantly lower before 2009. In contrast, for incidents causing damage, there was no substantial difference between the incidents when the pilot was warned and when they were not.

C. Engine - Feature¶

• Engine feature contains null values. These will be filled with the median, or the most frequent value.

Visualize the distribution, we'll remove null values.

Faceted plot, each plot corresponds to a different subset of the data based on the "Engine" column.¶

Plotting Probability Density Functions (PDFs) for 'Engines' based on the 'Damage' column¶

Violin plot¶

Observations:¶

1. NumberStruckActual: Most incidents had only 1 actual struck.
2. PeopleInjured - Here we see that in 13 events people were injured.
3. IsAircraftLarge? - Most incidents with no damage involved smaller aircraft. Large aircraft, have a higher count in the non-damaging incidents but are involved less in the damaging incidents.

Bi-variate analysis¶

Box Plot

The number of engines is not a factor to be considered for people getting ingured.

The number of engines did not appear to affect the number of people injured. While it seems that larger wildlife sizes had a greater impact on 1-engine aircraft, we know that in the incident where 6 people were injured, small wildlife was involved. Additionally, the number of incidents with injured people is not large enough to draw definitive conclusions (Fortunately, there have not been many dangerous incidents).

There does not seem to be a relation within how different amount of birds struck damage engines, as the incidents with high amounts of birds were not many.

Observations:¶

1. Most of the incidents involved 2 engine aircraft.
2. Of those incidents, most did not result in damages.
3. As for aircraft with 1, 3, and 4 engines, the difference between the resulting damage/no damage was not as great.
4. Interestingly, the violin plot shows that when the incident involved a 2-engine aircraft and resulted in damage, medium-sized wildlife caused similar damage than the large size.

Correlation¶

The data shows weak correlations overall, with some indications that 2-engine aircraft are involved in more incidents, and larger aircraft are less likely to experience damaging incidents.

Trend¶

The trend suggests that two-engine aircraft have been involved in a higher number of incidents.

Note: This could imply that two-engine aircraft are more prone to incidents, or it could be that there are simply more of these aircraft in existence, leading to a higher number of incidents.

D. ConditionsPrecipitation - Feature¶

Note that this feature contains only 2,015 non-null values, so the analysis is preliminary and should be interpreted with caution

Observations:¶

1. Although the sample data is limited, it provides a reasonable insight into how weather impacts bird-aircraft collision occurrences.2.
2. Most incidents occurred during "Rain", although most did not result in damages.
3. During "Rain, Snow" conditions, incidents were more likely to cause damage; however, these conditions accounted for only a small number of occurrences.

Trend¶

Rain increases the frequency of bird-aircraft collisions, but these incidents are generally less likely to cause damage. However, when rain and snow conditions are present, incidents are more likely to result in damage, although these conditions occur less frequently.

E. MakeModel - Feature¶

Observations:¶

1. Among all the models involved in bird-aircraft collisions, only a small number experienced repeated incidents.
2. Aircraft model B-737-700 (2-engine) had the most incidents.
3. As seen before, most of the models involved in incidents had 2 engines.

Trend¶

The trend indicates a strong association between the number of engines and the likelihood of involvement in collision.

Note: As stated before, this could imply that two-engine aircraft are more prone to incidents, or it could be that there are simply more of these aircraft in existence, leading to a higher number of incidents.

F. NumberStruckActual - Feature¶

Pie chart to visualize the distribution of the number of birds struck on a single incident¶

Multivariate analysis¶

Correlation Between NumberStruckActual, Damage, Engines, IsAircraftLarge? and PeopleInjured

The inference we can draw from this table is:

The heatmap shows there is little correlation between the considered factors.
There is some negative correlation between the number of Engines and Damage. As we've seen the 2-engine aircraft had the most incidents.
There is a very weak positive correlation between Engines and the number of PeopleInjured.

Observations:¶

1. Over 80% of incidents implied only one bird.
2. The vast majority of incidents where there was only one bird involved,did not cused damage.
3. Overall ost incidents did not caused damage.

Trend¶

The data show that most incidents implied only one bird.

G. FlightDate - Feature¶

Explore Trends Over Time: Yearly distribution of bird strike incidents. We need to group FlightDate column by year

Correlation Between Year and Damage¶

Observations¶

1. Although most of the incidents took place in 2009, the amount of incidents that caused damage has been very constant through the years.
2. From 2000 to 2008 the incidents gradually increased, before jumping steeply in 2011. After that incidents started decreasing.

Trend¶

The overall trend is upward. Data shows initial growth over 9 years, followed by a decrease over the next 3 years.

H. OriginState - Feature¶

Correlation Between OriginState and Damage for the top 30 incidents

Comments¶

1. California, Texas, and Florida are the states with more incidents compared to their countries.
2. It is fair to say that incidents have happened in all of U.S. territory.

Trend¶

The incidents show a localized upward trend in 3 key airports: California, Texas, and Florida.

I. Cost - Feature¶

Note: "CostAsFloat" column was added above, on: 5. Data Quality Assessment: Outliers for analysis.

Create a histogram to visualize the distribution of Cost

Observations¶

1. Although there was one very costly incident, most are not as expensive
2. The most costly incident resulted in 2 people being injured. And it was not a collision against a bird.

Trend¶

The most costly incident involved 2 people being injured, but it was not related to a bird collision, indicating that the severity and cost of incidents may not always correlate with bird strikes.

Although the dataset is highly usable and provides valuable insights, there are some gaps that limit a deeper understanding of the trends. For example, we observed an increase in incident occurrences over time, but it's unclear whether this is due to the growing number of planes in operation, urban expansion pushing into bird habitats, or potential changes in aircraft engine design.

While the data offers a strong foundation for analysis, it raises several questions that can only be answered with additional external data. Factors such as the number of aircraft, environmental changes, and even advancements in aviation technology could all influence the trends seen in this dataset. To fully comprehend the underlying causes and patterns, more comprehensive data will be needed to explore these variables in greater detail.

J. PeopleInjured - Feature¶

Exploring the incidents where people sustained injuries.

Visualizing correlation matrix heatmap.

Observations¶

1. Approximately 0.05% of incidents resulted in injuries to people.
2. There is a a weak correlation between "PeopleInjured" and these other factors: "NumberStruckActual", "Engines", "CostAsFloat".
3. Interestingly, the incident that resulted in the highest number of injuries involved small birds.

Correlation¶

The data shows little correlation between the fact that people got injured and other factors.

7. Limitations and Challenges¶